31st Edition | Radiology AI Under Attack?

Campbell Arnold
Feb 18
5 min read

“National Government Services does not consider automated detection and quantification of brain MRI imaging to be reasonable and necessary.”

— National Government Services Medical Policy Unit, Jan 22nd 2026

Welcome to Radiology Access! your biweekly newsletter on the people, research, and technology transforming global imaging access.

In this issue, we cover:

Medicare Contractor Proposes Non-Coverage For Quantitative Brain AI
Agentic AI: Radiology’s Next Leap, or Next Buzzword?
Microsoft begins operating PhysioMRI’s ODIN system

If you want to stay up-to-date with the latest in Radiology and AI, then don't forget to subscribe!

🚀 We’re Hiring at Subtle Medical!

Interested in building AI that actually ships into clinical practice and impacts patient care at scale? Subtle Medical is hiring US-based Research Scientists and Machine Learning Software Engineers to work on cutting-edge algorithms used by hospitals around the world.

If you’re excited about deep learning, medical imaging, and translating research into real products, I’d love to hear from you. Email me directly with your resume attached.

Medicare Contractor Proposes Non-Coverage For Quantitative Brain AI

CMS signals a higher bar for clinical utility, how will this reshape AI reimbursement?

A quiet policy proposal could have outsized consequences for the future of radiology AI. Earlier this month, a Medicare Administrative Contractor covering 10 states issued a draft Local Coverage Determination proposing non-coverage for automated detection and quantitative analysis of brain MRI. This includes services billed under CPT codes 0865T and 0866T, which were pioneered by Icometrix and ASNR in a push to establish reimbursement pathways for quantitative brain algorithms.

The draft LCD argues that existing evidence is insufficient to demonstrate improved patient outcomes in the Medicare population, emphasizing concerns around generalizability, dataset diversity, and downstream clinical impact. CMS outlines several key findings that underpin its non-coverage stance:

Automated brain MRI quantification remains “investigational” absent evidence linking results to changes in clinical management or outcomes.
FDA clearance alone is insufficient to justify Medicare payment.
Study populations and training datasets are limited, raising concerns about applicability to older, more heterogeneous Medicare patients.
Quantification without demonstrated decision impact does not meet coverage thresholds.
Physician interpretation must remain central, with AI positioned strictly as an adjunct.

For a category that has spent years transitioning from research into routine clinical workflows, this represents a significant potential setback. Private payer spillover is underway as well, with UnitedHealthcare announcing prior authorization requirements for CPT 0865T and 0866T beginning April 1, 2026. Together, these moves signal a tightening reimbursement environment for neuroimaging AI, which could influence investment decisions, product strategy, and clinical adoption.

Now is the moment for the radiology AI community to engage. CMS is accepting public comments on the proposed determination through March 8, 2026, with an open meeting scheduled for February 26, 2026.

Bottom line: Without stronger, outcome-driven evidence, quantitative brain MRI AI risks losing reimbursement momentum just as it was beginning to achieve real clinical adoption. Consider commenting or attending the meeting to help shape the final decision.

Agentic AI: Radiology’s Next Leap, or Next Buzzword?

As new AI systems emerge, radiologists face a choice, lead or risk losing influence.

Agentic is the latest AI buzzword, but is it actually taking hold in radiology, or is it mostly hype? Radiologists have been leaders in clinical AI adoption, yet most tools on the market today remain narrow, single-purpose devices rather than true collaborators or copilots. A new editorial in Radiology poses a provocative question: what if AI could reason, plan, and act across the entire radiology workflow, instead of simply flagging findings?

The authors describe agentic AI as a shift away from static algorithms toward systems that can decompose goals, coordinate multiple tools, and iteratively reason (albeit always under radiologist supervision). Unlike most FDA-cleared devices that perform a single, well-defined task, agentic systems could:

Integrate images, EHR data, prior studies, and clinical guidelines into a unified workflow.
Draft reports, prioritize cases, and manage follow-ups, not just detect abnormalities.
Adapt in real time, reflecting on intermediate results and course correcting.

While fully autonomous agentic systems aren’t being deployed anytime soon, the editorial outlines a realistic near-term vision for human-in-the-loop AI copilots with guardrails such as approval checkpoints, confidence thresholds, and audit trails. Still, clinical deployment of agentic systems poses major challenges, including clinical risk, regulatory uncertainty, workflow disruption, liability, and trust.

While radiologists have historically driven AI adoption, a growing share of physicians are using AI and deployment decisions are now being made by non-physicians as AI spreads across health systems. If radiologists want to remain the domain experts over AI in medicine, they must help shape how these agentic systems are designed, governed, and deployed. The window to lead is open, but it may not stay that way for long.

Bottom line: Agentic AI is not here yet, but if radiologists don’t actively shape how these systems are built, governed, and integrated, they risk losing influence over AI in healthcare.

Microsoft begins operating PhysioMRI’s ODIN system

Is mobile MRI moving from prototype to real-world use?

In a milestone moment, PhysioMRI has announced its first flagship client: Microsoft. Earlier this month, the Spanish portable MRI startup revealed that researchers at Microsoft Research were trained on and are now operating the ODIN system.

Unlike conventional MRI scanners, which rely on multi-ton high-field superconducting magnets, PhysioMRI’s approach embraces low-field, compact systems that are significantly lighter, easier to operate, and far more accessible. The ODIN device operates at a field strength of 87 mT, stands about 5.5 feet tall, and weighs just over 1,000 pounds. Its elliptical bore is 11x7.75 inches, making it primarily suited for extremity imaging, though head imaging may be possible for some patients. Earlier prototypes have already been used for extremity imaging at the MotoGP World Championship, demonstrating real-world applicability.

PhysioMRI’s broader NextMRI project is backed by the European Innovation Council to develop portable low-field scanners with mobile architectures that can plug into standard power sources. By securing a high-profile client like Microsoft early in its commercial portfolio, the company not only validates the practical utility of portable MRI but also signals growing interest from research and innovation leaders in exploring low-field use cases and data-centric approaches to imaging. As portable MRI systems gain traction in emergency medicine, neurology, and community care settings, this collaboration could catalyze broader adoption and spur further innovation in accessible, point-of-care diagnostic imaging.

Bottom line: The Microsoft partnership is a major milestone for PhysioMRI, does this signal growing interest in broader audience for making accessible, point-of-care imaging a reality?

Feedback

We’re eager to hear your thoughts as we continue to refine and improve RadAccess. Is there an article you expected to see but didn’t? Have suggestions for making the newsletter even better? Let us know! Reach out via email, LinkedIn, or X—we’d love to hear from you.

References

Disclaimer: There are no paid sponsors of this content. The opinions expressed are solely those of the newsletter authors, and do not necessarily reflect those of referenced works or companies.